How to explore: drag to rotate · scroll to zoom · hover for a quick label · click for full context · toggle the detail panel with the button ↗
The dissertation tests whether Soviet-era talking points on famine commemoration transform predictably when they cross register boundaries - secret memo → embassy/Novosti → domestic news → diaspora/scholarly print - and whether that transformation is dated and measurable (Montreal 1983 as anchor, post-1983 explosion as qualitative rewire). These 3D models make Lotman’s spatial vocabulary falsifiable: they read from the same open JSON as Figures 1-12.
This is a three-dimensional heatmap of operators. Each bar = one operator (scare quotes, drought frame, genocide commemoration…) in one document. Taller bar = more matches in that text. The “mountain range” on internal memo columns - especially scare quotes on spr. 1206 - drops to zero on the embassy column at the same operator row. Same three topics (famine, symposium, gathering); incompatible required readings. The gap is counted untranslatability.
Left → right = text surfaces from secret memo through outward Soviet lines to diaspora press. Front → back = operator type. Height = raw hit count O[operator, surface] from the corpus JSON. Colour = register family (internal blue, outward green, press orange, rival red).
There are no drawn arcs here - the comparison is spatial. When you mentally connect the tall internal scare-quote bar to the flat embassy bar on the same row, you are testing relay: did the operator travel outward? The answer is measurable: Mscareint = 48 vs 0 on embassy. Denial frames (drought, kulak, no artificial famine) rise on outward columns instead. That skew is the dissertation’s operator evidence for Lotman’s untranslatability - not a metaphor, a counted gap.
Thesis link: The tensor makes Lotman’s ‘untranslatability’ countable: internal columns spike on scare quotes and gathering downgrade; embassy and Novosti columns stay flat on those operators while rising on denial frames. That spatial gap is the dissertation’s operator-skew evidence - same three topics, incompatible required readings.
Try it: click the tallest scare-quote bar on spr. 1206 - read “Why it matters to the dissertation.” Then click the embassy column on the same operator row and compare height. Pair with L1 arcs: thematic relay carries denial, not scare quotes.
Each panel opens one step on the way from the corpus counts to the bars, with the equation it uses.
An operator is a wording habit the project tracks, such as scare quotes, calling the symposium a "gathering", or answering famine talk with drought. A surface is one document. The project's operator detectors in the thesis lab count the matches. Ten operators across thirteen surfaces give 130 cells, and 37 of them are above zero.
Height is proportional to the count, scaled so the largest cell (19 scare quotes on spr. 1208) stands 8.15 units tall. The extra 0.15 keeps a single match visible. Cells with no matches are not drawn, so an empty spot in the grid is a real zero. Surfaces run left to right from the secret memos to the diaspora press, operators run front to back, and colour is the register family: internal blue, outward green, press orange, rival red.
The comparison the figure is built for: scare quotes across the three internal memos against the embassy release in the same row.
Nothing in the data moves. Turn the model spins the bars slowly so rows and columns hidden behind tall towers come into view. Dragging still works while it turns.
Generated in the thesis lab by build_lotman_3d_visualizations.py from the corpus JSON and copied here on sync. The data object at the bottom of the page holds all 130 cells. The 2D version is the honest plain heatmap. This section is kept in explain/fig-lotman-operator-tensor-3d.html in the site repo and added back on every sync.
The secret memos are full of scare quotes, 48 across three of them. The embassy release, written for foreign readers about five weeks after the first memo, has none. It explains the famine with drought and kulak sabotage instead, habits that never appear in the three memos. The same topic carries a completely different set of habits depending on the audience.
This is the counted version of what Lotman's boundary does: the memo's way of talking about the event does not cross into the public text, even though the event does. The dissertation calls that gap untranslatable residue and counts it rather than asserting it.
It also shows why language-model fingerprints can't stand in for reading. Fingerprints group texts by topic, so the memo and the embassy release land close together. Scare quotes and the habits in this chart are exactly what they tend to miss. A retrieval system built on fingerprints alone will hand you a secret memo when you asked for the newspaper.
Robust where it is simple. The scare-quote and denial-frame counts are large, one-sided and easy to check by hand. The weak point is the detectors themselves. They count patterns, so a stray quotation mark in older idiom registers as a scare quote, which is where the 1 to 3 hits on the domestic papers come from. Small counts on single documents shouldn't carry an argument.
Suggestions: a blind human check of a sample of matches for each operator, so every detector has a known error rate; more documents per register, especially domestic newspapers once the East View clips arrive; and a second person checking the operator definitions. With several documents per cell, simple tests of whether a row really differs between registers become possible.